RFR: JDK-8160571: Bootcycle builds still broken with server jvm on Windows 32bit

[David Holmes]

On 30/06/2016 3:16 AM, Erik Joelsson wrote:
> In JDK-8160285 I tried to fix the bootcycle build on 32bit windows when
> only the server jvm is being built. It appears I only verified the fix
> on a release build. In my JPRT job I forgot to enable only the server
> jvm so it ran with client.
>
> We are now faced with more similar failures in the fastdebug build. The
> javac server process is set to use mx1500M and fails all the time in
> that configuration. The JAVA_FLAGS_BIG config fails intermittently with
> not being able to allocate enough memory.
>
> My suggested fix is to further reduce the maximum amount of heap from
> 1100 to 1024 to avoid the intermittent failures and to replace the
> override of the "javac server jvm args" with the same set of args as
> have already been calculated for "java big". I have run this
> configuration through JPRT with client disabled. I have also run several
> builds locally without failure.

Seems okay, but I hope it will be big enough. As mentioned offline the 
switch from client to server has also affected GC configuration which 
can create many threads and consume much additional C heap. The notions 
of "big" and "small" java may need further consideration.

Thanks,
David

> Bug: https://bugs.openjdk.java.net/browse/JDK-8160571
> Webrev: http://cr.openjdk.java.net/~erikj/8160571/webrev.top.01/
>
> /Erik